Self-help psychological intervention for young individuals during the post-COVID-19 era: development of a PST chatbot using GPT-4.

Abstract

Introduction: The COVID-19 pandemic has exacerbated psychological stress among young people, with some survivors experiencing persistent mental distress, thus creating an urgent need for accessible psychological intervention tools. To help young people affected by COVID-19 recover and achieve balanced mental health in the post-pandemic era, this study developed an online self-help psychological intervention chatbot to supplement existing mental health resources.

Methods: We utilized prompt engineering techniques to construct a chatbot proficient in Problem-Solving Therapy (PST) based on the large language model GPT-4. Subsequently, 7 master's students majoring in psychological counseling were recruited for a pre-test of the chatbot, and 100 young people who had contracted COVID-19 were selected for a formal user experiment to evaluate its effectiveness.

Results: The pre-test results indicated that the chatbot followed the core steps of PST during interactions with users and was helpful in problem-solving. The formal experiment showed that the experimental group scored significantly higher than the control group in the dimensions of problem awareness [t (88.31) = 3.14, p = 0.002] and problem-solving [t (98) = 3.34, p = 0.001], but there was no significant difference between the two groups in the dimension of relationship quality [t (91.23) = 1.07, p = 0.286]. In addition, no significant differences were found in the evaluation based on gender or the presence of post-COVID-19 symptoms, indicating that the chatbot has a certain degree of universal applicability.

Conclusions: These findings support the application of the PST chatbot in post-COVID-19 era psychological interventions, particularly in assisting users with identifying problems and exploring solutions. Although the chatbot did not achieve significant improvement in human-computer relationship quality, its general acceptability and broad applicability demonstrate great potential in the field of mental health, highlighting the value of large language models in promoting self-help mental health interventions as a supplementary tool to existing resources.